Higgs boson production in association with a top quark pair: tth(bb) with the matrix element method

Transcription

1 Higgs boson production in association with a top quark pair: tth(bb) with the matrix element method Joosep Pata PSI PhD seminar In collaboration with L. Bianchini, L. Caminada, F. Canelli, G. Dissertori, G. Kasieczka, D. Salerno

2 H ττ tagged µ =.83 ±.35 µ =.9 ±.28 tth tagged H bb tagged µ = 2.75 ±.99 µ =.84 ±.44 What we know about H(25) 2 ATLAS-CONF Best fit σ/σsm CMS-HIG-4-9 Combined µ =. ±.4 - Be fb (8 TeV) + 5. fb (7 TeV) mh = 25 GeV CMS H γ γ (untagged) H γ γ (VBF tag) psm =.84 H γ γ (VH tag) H γ γ (tth tag) H ZZ (/ jet) H ZZ (2 jets) H WW (/ jet) H WW (VBF tag) H WW (VH tag) H WW (tth tag) H ττ (/ jet) H ττ (VBF tag) H ττ (VH tag) H ττ (tth tag) H bb (VH tag) H bb (tth tag) Best fit σ/σsm Figuredata 4: Values of the best-fit for the combined analysis SM boson, H(25) coupling mostly consistent with SMs/s Higgs butoverall small differences could be very (solid vert interesting! separate combinations grouped by production mode tag, predominant decay mo - The s/ssm on ratio denotes the section times phase! the relevant ~5% uncertainty top Yukawa with 3production fb in Run II: cross reaching the precision branchi relative to the SM expectation. 2 The vertical band shows the overall s/ssm unce

3 Coupling to fermions Verify EWSB mass generation by comparing SM H decay to fermions to measurements. Direct evidence down-type (b, tau), indirect to up-type (top loops) (PRD) Data Background (µ=.4) Background (µ=) 3 2 Entries /.25 Events / bin arxiv: (JHEP) H (25) ττ ( µ=.4) 7 CMS s = 7TeV, L = 5. fb - s = 8TeV, L = 8.9 fb pp VH; H bb Data MC(B) H ττ - log(s / B) Data MC(S+B) s = 7 TeV, 4.5 fb =.64-3 χ2/ dof =.98-3 χ2/ dof.5 s = 8 TeV, 2.3 fb- VH(bb ) 25 GeV bb ATLAS -4 indirect evidence for coupling to tops through loop diagrams Background Background uncert. -4 VH 3 2 Data - 6 H (25) ττ ( µ=) ττ log(s/b)

4 What to look for? H(25) H(25) near-ideal for bb statistics-wise, but a very complicated final state. 4

5 Final state 4 Event reconstruction and selection b t g ` W+ t H `+ b b q, ` t g W t q, ` b ng-order Feynman diagram for tth production, illustrating the two top-quark y channels considered here, and the H! bb decay mode for which the anal. -2 leptons, 4-6 jets (4b), METindividual objects (electrons, muons, and ommon set of criteria for selecting cribed below. 5 J. Pata channel, the data were recorded with triggers requiring the presence of either

6 Reconstruction and selection 6 Run I Require -2 isolated e/mu multijet QCD, but also fully-hadronic tth(bb). Events (EPJ C) CMS SL Data Bkg. unc. tth (25) x tt + lf tt + cc tt + b tt + b b EWK Single t tt + W,Z VV 3 4 or more central jets pt > 3(2) GeV, b-jets based on impact parameter and SV properties. single top, EWK. 2 Data/Bkg b-tagging for tt+heavy vs. light: Nb, likelihood discriminant tt+light fb- (8 TeV) Jet multiplicity.5.5 Events 9.5 fb- (8 TeV) CMS SL Data Bkg. unc. tth (25) x 5 tt + lf tt + cc tt + b tt + b b EWK

7 top quark background Bevilacqua et al 24 vs. LHC XSWG tth: 58.5 [fb] ± ~% ttbb uncertainties cover expected signal! Modelling uncertainties of tt+bb (gluon splitting). tt + light vs. bb F-discriminant top quark pair background tt+bb tt+h(bb) tt+light tt+bb vs. tt+h(bb) combined fit in σ(tth) vs. σ(ttbb)/σ(ttjj) space! 7

8 is rigorous for W!EWK ud(s) decays, whereas it is onlyewk approxima Single t Single t tt + W,Z tt + W,Z since the CSV discriminator pdf for charm quarks differs VV VV with resp can be extended to the case of SL events with five jets, or DL event considering that in both cases four of the jets are associated with h the remaining jets with light-flavour partons. The likelihood under f (~ x tt+lf), is given by Eq. () after swapping f hf for f lf. The varia then defined as the likelihood ratio 4 3 b-tagging: tt+heavy vs. tt+light Jet multiplicity tracks Jet multiplicity ~x tt+hf) f ( F (~x ) =. f (~x tt+hf) + f (~x tt+lf) Data/Bkg Data/Bkg 4.5 jet The with six jets is shown in Fig. 2 (b cone CMS distribution of F for SL events CMS.5 Events 9.5 fb- (8 TeV) 6 SL Data Bkg. unc. tth (25) x 5 tt + lf tt + cc tt + b tt + b b EWK Single t tt + W,Z VV 4 tt + lf tt + cc tt + b tt + b b EWK Single t tt + W,Z VV Data Bkg. unc. tth (25) x p 4 Multiplicity of b-tagged jets (CSVM) Data/Bkg Data/Bkg PV SL In the following, events are retained if F Iis larger than a threshold Run in multi-jet,.85 and.97, depending on thechannel and jet multiplicity. The combine classified as high-purity (low-purity) if F is larger (smaller) than a v probabilities:.. The low-purity categories serve as control regions for tt+lf jet p(jet b) uncertainty. ~ mm The high-purity catego several SV sources of systematic & p(jet non-b) b meson IP 5 track 9.5 fb- (8 TeV) (EPJ C) Events F.5 Figure 2: Top row: distribution of the jet multiplicity in (left) single-lepton and (right) dilep-

9 b-tagging in Run II Studying a new super-mva for b vs. light. Combines different vertexing algorithms, likelihood-based jet probability, soft lepton taggers. CMS preliminary Run II simulation c mistag, high-eff b-tagging improved by ~2-3% (rel.). new MVA 9

10 tt+bb vs. tt+h(bb) We could reconstruct H( bb) system invariant mass, but how to choose the right b-jet candidates? ATLAS Preliminary H( bb) smeared out due Data 22 tth (25) tt+v tt+bb to combinatorics, cannot non-t t tt+cc Total unc. tt+light observe Higgs peak. tth (25) norm Events / 22 GeV ATLAS Preliminary - L dt = 2.3 fb s = 8 TeV Single lepton 6 j, 4 b Data 22 tth (25) tt+v non-tt Total unc. ATLAS-CONF-24- tt+bb tt+cc tt+light tth (25) norm Need to exploit H resonance properties experimentally. 8 Data / Pred Run I [GeV] min R m bb

11 Main problems tt+jets without H dwarfs the signal by several orders of magnitude. complicated final state: multi-jet, missing energy

12 Extracting the signal 4 Reconstructed event probability density value from theoretical models: Matrix Element Method (MEM), full use of kinematics in LO (EPJ C) Events Unknown, poorly measured quantities integrated out directly. Run I 4 CMS 9.5 fb- (8 TeV) Ph/l<.5 Ph/l>.5 Data Bkg. unc. tth (25) x SL Cat- (H) tth (25) tt + lf tt + cc tt + b tt + bb Single top tt + W,Z 2 combinatorics for associating the multi-particle final state to the theoretical model. 5 5 Data/Bkg Ps/b CMS fb- (8 TeV) Events High-level features: neural networks (NN) or boosted decision tree (BDT) -> exploit best simulation (NLO), but little physical insight, need high statistics. 8 Ph/l<.5 Ph/l>.5 Data Bkg. unc. tth (25) x SL Cat-3 (H) tth (25) tt + lf tt + cc tt + b tt + bb Single top tt + W,Z

13 from ected by beyond-standard model (BSM) particles. In order to that of Cat- by the inversion of the dijet m wa coupling from a possible BSM contribution, a direct tion mea-assumes that one of the quarks from the W b integration on the right-hand side of Eq. (3) is ext d. This can be achieved by measuring observables that probe quark. The other untagged jet(s) is (a n with the Higgs boson already at the tree-level. Theconstructed proenter the calculation of w(~y H). The total numbe gs boson in association with a top-quark pair (tth) provides multiplicity of untagged jets eligible to originate e. A sample of tree-level Feynman diagrams contributing to third category (Cat-3), exactly five jets are req tth is shown in Fig. (left and centre). The inclusive the nexttion is again assumed. In the DL channel, only o s section is about 3 fb iny pp-collisions at a centre-of-mass measured event properties that each of the four bottom quarks in the decay boson mass (mh ) of 25 GeV [5 24], which is approximately - true parton-level quantities Finally, two event discriminants, denoted by Ps than the cross section forx Higgs boson production via gluon only information from the event kinematics and to separate the signal from the background; th H H g g t b b tagging, thus providing a handle to distingu likelihood ratio They are components of the tt+jets background. Matrix Element Method H or b t Ps/b w(~y tth) = w(~y tth) + ks/b w(~y tt+bb) and where the functions f (~x tt+hf) and f (~x tt+lf) ar g g t t only to the jet-quark associations considered in th kh/l in the denominators are positive constants th grams contributing to the partonic processes: (left) qq! tth, The joint distribution of the ( Ps/b, Ph/l ) discrimi! tt+bb. likelihood fit to search for events resulting from p b satisfyathe bconstraint Ps/ a two discriminants ed pp collision data at s =.96 TeV collected by the CDF ex2simulated samples, the distributions of Ps/b and [25]. Searches for tth production at the LHC have previously former, cay modes of the Higgs boson [26, 27]. The first combination of which carries the largest sensitivity to t equal-sized bins. The coefficient ks/b appearing i es has been published by the CMS Collaboration based on the the relative normalisation between w(~y tth) and 3 nd 8 TeV [28]. Assuming SM branching fractions, the results of parton level reconstruction level f(y H, λ) = /σ(h, λ) dx dx dx Φ(x,x ) PDF ℳ(X H, λ) W(Y,X) ME LO exp. resolution

14 Reconstruction categories W qq in one light quark not Categories Categories reconstructed acceptance hypo Define categories and do event interpretation Define categories and do event interpretation Reconstructed event b Cat. b ucat. 2 b2 u b2 W qq? b3 b3 W qq Interpretation b4 Categories μ Integration gluon? Friday, October 4, 3 Friday, October 4, 3 W qq Discriminant P u2 b4 μ! (M(ubi) [65,95] GeV & csvbi<.9) 36 SL Friday, October 4, 3 Categories u2 μ Define categories and do event interpretation M(upermutation M(uu2categories ) [6,]and GeV u2) [6,] GeV Define do event interpretation 24 permutation b b Cat. 6 b2 Cat. 4 b2 u μ b3 b4 u2 4 b3 b4 μ 2 permutation Friday, October 4, 3 DL 38 2 permutation

15 ME is just a discriminator, no need to be fully theoretically sound. An approximation already does a very good job. Preliminary CMS simulation Run II bg-like signal-like 5

16 9.5 fb- (8 TeV) Expected ± σ Expected ± 2σ Expected (sig. inj.) Observed CMS SL CMS SL Results from LHC run I DL DL Combined Combined CMS tth(bb) with MEM 2/fb: μ <95%4.2 (3.3) CL limit on µ = σ/ σ obs 95%CL -4 6 Events SM at mh = 25 GeV (EPJ C) Run I CMS Best f 9.5 fb- (8 TeV) Data Signal (µ = ) Background Bkg. unc. 9 Summary3 CMS with NN 5+5/fb: μ < 5.8 (5.2) -2 ATLAS MEM(SL) + NN 2/fb: μ < 3.4 (2.2) in Run II, first limits with ~/fb. 9.5 fb- (8 TeV) SL 9.5 fb- (8 TeV) CMS SL DL DL Combined Combined 95% CL limit on µ = σ/ σ SM at mh = 25 GeV Best fit µ = σ/ σ at mh = 25 GeV SM Events log(s/b).5.5 Figure 5: (top left) Observed 95% CL UL on µ are compared to the medi Data (µ = )background-only and the signal-plus-background hypotheses. dersignal the 6 Background 9.5 fb- (8 TeV) CMS Data/Bkg Expected ± σ Expected ± 2σ Expected (sig. inj.) Observed CMS 2 Bkg. unc.

17 Ongoing work Improve MEM categorization: less constrained hypotheses - more separation at the cost of CPU time. Combination of MEM discrimination with NLO-based machine learning. Jet substructure can improve already existing limits: top tagging [53.592], higgs tagging of fat jets [ ], direct integration with MEM/MVA. Deploy MEM in additional topologies: H(ττ), fully-hadronic tth(bb). Fully exploit and improve b-tagging to constrain tt+bb/cc/light, quarkgluon discrimination. 7

18 Top quark physics Couplings Moving to LHC Run II By far the strongest of the H gg-dominated (ratio to W.J.σ(ttH) Stirling = 623 fb at 4 TeVWJS23 complement golden bb Cnal state w tth ~.585 pb (3.9x) favoured over tt+jets ~832 pb (3.3x) With 3/fb, coupling scale factor κt uncertainty.5. luminosity ratio ratios of LHC parton luminosities: 3 TeV / 8 TeV moderate excess observed in ggnb: _ Σqq qgcrucial to follow up with Run II dat Projections for yt evolution bas. saturated systematics MSTW28NLO 2. theory reduced by ½, exp. scaling w MX (GeV) Jet, lepton spectra harder in pt, higher multiplicity, increased signal acceptance. Run II But crucial to keep b-tagging performance with high pile-up, exploit jet substructure. 8 arxiv:37.735

19 Summary & outlook tth(bb) MEM proof of concept at 8 TeV: μ < %. Need the best possible interpretation of data : matrix element method natural for high jet multiplicities, complements machine learning. tt+bb NLO crucial, MEM naturally suited for systematically dominated (LHC 3/fb). Extend to additional reco. hypotheses, use of jet substructure information. LHC has only started to speak up about the Higgs, Run II will be crucial! 9

20 Backup 2

21 ME details OpenLoops-based. Verified against MadWeight at discriminator-level. - CPU-sec / event. Integration via VEGAS, PDF using LHAPDF. Detector effects encoded in transfer functions. assumptions No spin correlations. Leading order only - no pt etc. Narrow-width t, W, H. Only ggh. Require b-tagging of b- quarks (F-discriminant). 2

22 Expected rates S/ B ~ 2% after lepton and jet selection. CMS preliminary Validate analysis in different jet / b-tag categories, combined fit. Run II simulation MEM hypotheses optimized per category. signal-like, one missing jet 22 signal-like, fully reconstructed

23 Uncertainties Differential (shape) and inclusive (normalization) cross-sections affected. Fit μ within uncertainties. CL based on profile likelihood q(μ) properties tt+jets modelling: tt+heavy flavour cross-section (data 5-2%), renormalization, factorization and resummation scale and functional form, shower recoil, PDF choice, MPI, FSR, top quark and top pair kinematics. Jet energy scale and resolution: e parameters. Based on the asymp q(µ) = 2 ln L µ, ˆq µ /L ˆµ, ˆq, he best-fit value for obtained wh B-tagging: discriminator distributions, not only total efficiencies need to be modelled well -> differential corrections. Theory uncertainties on tt+h and tt+jets discriminant shape: crucial to describe gluon splitting. 23

24 Data & models ATLAS and CMS: 5+2 fb - of 7 and 8 TeV collision data with lepton triggers -> 3.9 x 3 tth events [CERN-23-4], 55 (65) % gg->h at (N)LO in 8 TeV, more in 3 TeV signal: LO->NLO simulation for tt+h, corrected to NNLO + leading log, Pythia (amc@nlo at 3 TeV) tt+bb: LO -> NLO, significant theoretical uncertainties, combine predictions from multiple generators, NLO recently available with Sherpa +OL tt+cc, tt+light: LO (MG5), empirical corrections to top-antitop system and top quark kinematics. minor: single-top: NLO (amc@nlo), W/Z+jets, diboson 24

25 First look at 3 TeV simulation Preliminary CMS simulation Preliminary CMS simulation 25

26 Case study: tth( bb) 4 4 Event reconstruct H(25) decays to b-quarks most abundant. t g b ` W+ `+ b t H b q, ` t CMS 2 simulation preliminary, s = 7 TeV c jet efficiency udsg jet efficiency g Multi-jet events with high bw t q, ` quark multiplicity are a prolific b CMS PAS BTV--4 source of tth. Figure : A leading-order Feynman diagram for tth production, illustrating th CM pair system decay channels considered here, and the H! bb decay mode fo TCHE TCHP ysis is optimized. SSVHE The experiments are good at nels, we use a common set of criteria for selecting individual objects (electr jets) which is described below. detecting jets from b-quarks: In the lepton+jets channel, the data were recorded with triggers requiring the p a single bmuon or electron. The trigger muon candidate was required to be iso -2% udsg fake rate with 7% activity in the event and to have p > 24 GeV for both the 2 and 22 data-t 2, the trigger electron candidate was required to have transverse energy E quark efficiency. be produced in association with at least three jets with p > 3 GeV, whereas - SSVHP JP JBP CSV T T T - electron trigger with minimum ET threshold of 27 GeV was used. In the dilep -3 m -4 data were recorded with triggers requiring any combination of electrons and d with pt > 7 GeV and another with pt > 8 GeV. The offline object selection b jet efficiency 26designed to select events in the plateau of the trigger efficiency turn-on curve Figure 7: Performance curves obtained from simulation f